Implementing Data Capture for ML Observability and Drift Detection

Explore the implementation of data capture for machine learning observability and drift detection in this 24-minute conference talk by Pushkar Garg. Dive into the complexities of modern ML systems, including data pipelines and transformations across multiple layers such as data warehouses and feature stores. Learn about the crucial role of ML observability in productionizing models and the importance of efficient data capture at prediction endpoints. Discover Pushkar's experience in coding an in-memory buffer for data capture and the lessons learned during the process. Gain insights into how downstream monitoring jobs utilize data capture logs to complete the ML observability loop. Benefit from Pushkar's decade-long expertise in machine learning, artificial intelligence, and platform building for training and deploying models.